Modeling Mechanical Properties of Aluminum Composite Produced Using Stir Casting Method

TL;DR

This study developed an ANN-based model to predict the mechanical properties of aluminum composite materials produced by stir casting, using extensive experimental data to accurately capture complex nonlinear relationships.

Contribution

The paper introduces a neural network model trained on experimental data to predict multiple mechanical properties of aluminum composites, addressing the challenge of modeling their nonlinear behavior.

Findings

ANN model achieved 2-7% error in predictions

Model accurately predicts tensile strength, elongation, hardness, wear resistance

Extensive experimental data supports model training and validation

Abstract

ANN (Artificial Neural Networks) modeling methodology was adopted for predicting mechanical properties of aluminum cast composite materials. For this purpose aluminum alloy were developed using conventional foundry method. The composite materials have complex nature which posses the nonlinear relationship among heat treatment, processing parameters, and composition and affects their mechanical properties. These nonlinear relation ships with properties can more efficiently be modeled by ANNs. Neural networks modeling needs sufficient data base consisting of mechanical properties, chemical composition and processing parameters. Such data base is not available for modeling. Therefore, a large range of experimental work was carried out for the development of aluminum composite materials. Alloys containing Cu, Mg and Zn as matrix were reinforced with 1- 15% Al2O3 particles using stir casting method. Alloys composites were cast in a metal mold. More than eighty standard samples were prepared for tensile tests. Sixty samples were given solution treatments at 580oC for half an hour and tempered at 120oC for 24 hours. The samples were characterized to investigate mechanical properties using Scanning Electron Microscope, X-Ray Spectrometer, Optical Metallurgical Microscope, Vickers Hardness, Universal Testing Machine and Abrasive Wear Testing Machine. A MLP (Multilayer Perceptron) feedforward was developed and used for modeling purpose. Training, testing and validation of the model were carried out using back propagation learning algorithm. The modeling results show that an architecture of 14 inputs with 9 hidden neurons and 4 outputs which includes the tensile strength, elongation, hardness and abrasive wear resistance gives reasonably accurate results with an error within the range of 2-7 % in training, testing and validation.